Anim Cogn
DOI 10.1007/s10071-014-0811-7
ORIGINAL PAPER
Comparing supervised learning methods for classifying sex, age,
context and individual Mudi dogs from barking
Ana Larrañaga • Concha Bielza • Péter Pongrácz
Tamás Faragó • Anna Bálint • Pedro Larrañaga
•
Received: 5 June 2013 / Revised: 22 September 2014 / Accepted: 22 September 2014
Ó Springer-Verlag Berlin Heidelberg 2014
Abstract Barking is perhaps the most characteristic form
of vocalization in dogs; however, very little is known about
its role in the intraspecific communication of this species.
Besides the obvious need for ethological research, both in
the field and in the laboratory, the possible information
content of barks can also be explored by computerized
acoustic analyses. This study compares four different
supervised learning methods (naive Bayes, classification
trees, k-nearest neighbors and logistic regression) combined with three strategies for selecting variables (all
variables, filter and wrapper feature subset selections) to
classify Mudi dogs by sex, age, context and individual
from their barks. The classification accuracy of the models
obtained was estimated by means of K-fold cross-validation. Percentages of correct classifications were 85.13 %
for determining sex, 80.25 % for predicting age (recodified
as young, adult and old), 55.50 % for classifying contexts
(seven situations) and 67.63 % for recognizing individuals
(8 dogs), so the results are encouraging. The best-performing method was k-nearest neighbors following a
Electronic supplementary material The online version of this
article (doi:10.1007/s10071-014-0811-7) contains supplementary
material, which is available to authorized users.
A. Larrañaga
Student at the Universidad Alfonso X El Sabio, Av. Universidad,
1, 28691 Villanueva de la Cañada, Madrid, Spain
C. Bielza P. Larrañaga (&)
Computational Intelligence Group, Universidad Politecnica de
Madrid, Campus de Montegancedo, 28660 Boadilla del Monte,
Madrid, Spain
e-mail: pedro.larranaga@fi.upm.es
P. Pongrácz T. Faragó A. Bálint
Department of Ethology, Biological Institute, Eötvös Loránd
University, 1117 Pázmány Péter sétány 1/c, Budapest, Hungary
wrapper feature selection approach. The results for classifying contexts and recognizing individual dogs were better
with this method than they were for other approaches
reported in the specialized literature. This is the first time
that the sex and age of domestic dogs have been predicted
with the help of sound analysis. This study shows that dog
barks carry ample information regarding the caller’s
indexical features. Our computerized analysis provides
indirect proof that barks may serve as an important source
of information for dogs as well.
Keywords Mudi dog barks Acoustic communication Feature subset selection Machine learning Supervised
classification K-fold cross-validation
Introduction
Canine communication (including dog–human communication) has become a well-studied topic among ethologists
in the last decade. Most efforts have focused on how and to
what extent dogs are able to understand different forms of
human communication, through visual gestures (Reid
2009), voice recognition (Adachi et al. 2007), acoustic
signals for ceasing or intensifying their activity (McConnell and Baylis 1985; McConnell 1990), and ostensive
signals (Téglás et al. 2012). However, it has also been
found that dogs can get their message across to humans, for
example, by turning their head or alternating their gaze
between the human and their target (Miklósi et al. 2000),
and that dogs can emulate other behavioral forms so as to
convey feelings, of guilt for example, in an appropriate
situation (Hecht et al. 2012).
Unlike taxon-specific chemical and visual communication (Meints et al. 2010; Wan et al. 2012), acoustic signals
123
Anim Cogn
are regarded as highly conservative and uniformly constructed within such broad groups of animals as avian and
mammalian species. Morton (1977) provided a set of socalled motivation-structural rules to explain this point.
According to his theory, the quality of the sound (pitch,
tonality) strongly depends on the physical (anatomical)
constraints of the animal’s voice-producing tract, which in
turn depends on the physical features of the animal itself
(size, for example). Stronger, larger specimens within a
species will usually be the dominant, aggressive animals
and smaller, younger individuals are usually the subordinates. Thus, the typical vocalizations (low pitched, broadband, noisy) emitted by the larger, more aggressive
individuals, for example, could, according to Morton,
evolve into the trademarks of agonistic inner states. Similarly, the typical vocal features of a smaller, subordinate
animal (high pitched, narrow band, tonal) could project the
lack of aggressive intent communicative meaning.
Dogs have a rich vocal repertoire, see (Cohen and Fox
1976; Tembrock 1976; Yeon 2007), like other closely
related wild members of the Canidae family. The ethological analysis of the possible functions of canine vocalizations has so far provided data about the individual-specific
content of wolf howls (Mazzini et al. 2013; Root-Gutteridge
et al. 2013), the indexical content of dog growls, related to
the caller’s body size (Taylor et al. 2008, 2010; Faragó et al.
2010a; Bálint et al. 2013), and the context-specific content
of dog growls (Faragó et al. 2010b; Taylor et al. 2009).
However, even though barking is considered to be the most
characteristic form of dog vocalization, exceeding the barks
of wolves and coyotes both in its frequency of occurrence
and variability (Cohen and Fox 1976), the functional aspects
of dog barks are surprisingly little known. The theoretical
framework for the information content and evolution of
barking in the dog involves very different assumptions,
ranging from the theory that it is a non-communicative
byproduct of domestication (Coppinger and Feinstein
1991), through the low-information level mobbing signal
theory (Lord et al. 2000), to the context-specific information
source theory (Feddersen-Petersen 2000; Yin 2002; Pongrácz et al. 2010). As dogs are the oldest domesticated
companions of humans (Druzhkova et al. 2013), dog barking may have acquired a ’new target audience’ in humans
during the many 1,000 years of coexistence. A possible
indirect proof of this is a series of playback experiments
which showed that humans are able to correctly categorize
barks according to their contexts (Pongrácz et al. 2005). As
for contextual content, human listeners also had consistent
opinions about the inner state of the barking dogs, and the
acoustic analysis of the barks revealed that humans base
their decision on the kinds of acoustic parameters of the
barks that were expected on the basis of Morton’s theory
(Pongrácz et al. 2006). Besides the pitch and the harmonic-
123
to-noise ratio, however, it was found that the inter-bark
interval (or ‘pulsing’) of the barks is also important when
assessing the inner state of the barking dog.
Although there are convincing empirical demonstrations
that dog barks show acoustic features that are seemingly
context specific (Yin 2002; Pongrácz et al. 2005), and we
have also learned that humans can decipher information from
dog barks regarding the context of vocalization and the inner
state of the animal, it is less well understood whether dog
barks carry an equally rich (or even richer) content of
information for another dog. Until now, there have been only
a few experiments with dogs as subjects which revealed that
dog barks do carry individual-specific cues. One used a
habituation–dishabituation paradigm (Maros et al. 2008;
Molnár et al. 2009), and the other was a computerized bark
analysis study (Molnár et al. 2008). These results raise the
question of whether dog barks carry a much wider set of
information about the vocalizing animal than humans are
able to decipher. Another intriguing problem is which
acoustic parameters could be responsible for the finer details
of the information content of dog barks. Based on the vast
literature of vocalization-based sex and individual recognition in other species, e.g., African wild dog, Lycaon pictus
(Hartwig 2005); white-faced whistling duck, Dendrocygna
viduata (Volodin et al. 2005); or Wied’s black-tufted-ear
marmosets, Callithrix kuhlii (Smith et al. 2009), one might
expect dog barks to also carry specific cues of the caller’s
individual features, such as sex and age, for example. There
are, however, considerable obstacles in testing such subtle
pieces of information using classical techniques (i.e., playback). Fortunately, the current age of computer-based
methods opens up the possibility for analyzing and testing
lots of sound samples with the help of artificial intelligence.
Machine learning techniques have been used in behavioral research on acoustic signals for a wide range of
species, see Table 1. For dolphins, artificial neural networks have been applied to model dolphin sonar, specifically for discriminating differences in the wall thickness of
cylinders using time and frequency information from the
echoes (Au et al. 1995). Also, support vector machines and
quadratic discriminant function analysis have been used to
classify fish species according to their echoes using a
dolphin-emulating sonar system (Yovel and Au 2010), and
Gaussian mixture models and support vector machines
have been employed to classify echolocation clicks from
three species of odontocetes (Roch et al. 2008). Differentiation of categories or graded barks in mother-calf vocal
communication in Atlantic walrus have been analyzed with
artificial neural networks and discriminant functions
(Charrier et al. 2010). Frog song identification to recognize
frog species has been carried out with k-nearest neighbor
classifiers and support vector machines (Hunag et al. 2009).
Linear discriminant analysis, decision tree and support
Anim Cogn
Table 1 Examples of machine learning technique usage from acoustic signals for different species with different aims
Animal
Aim
Technique
Reference
Dolphin
Discriminate cylinder thickness
ANN
Au et al. (1995)
Classify fish species
SVM, quadratic DFA
Yovel and Au (2010)
Odontocete
Classify echolocation clicks
GMM, SVM
Roch et al. (2008)
Walrus
Classify barks in mother-calf communication
ANN, DFA
Charrier et al. (2010)
Frog
Classify species
kNN, SVM
Hunag et al. (2009)
Linear DFA, trees, SVM
Acevedo et al. (2009)
Bird
Classify species
Linear DFA, trees, SVM
Acevedo et al. (2009)
Recognize individuals
GMM
Cheng et al. (2010)
Bat
Classify species
ANN, DFA
Parsons (2001), Parsons and Jones (2000)
Trees
Adams et al. (2010)
Random forests, SVM
ANN, DFA, kNN
Armitage and Ober (2010)
Britzke et al. (2011)
Cricket, grasshopper
Classify species
ANN
Chesmore (2001)
Marmot
Classify identity, age and sex
DFA
Blumstein and Munos (2005)
Suricate
Predict predator type
DFA
Manser et al. (2002)
African elephant
Classify vocalization type
HMM
Clemins (2005)
Classify contexts
HMM
Clemins (2005)
Recognize individuals
HMM
Clemins (2005)
Female elephant
Classify rumbles by oestrous cycle phase
HMM
Clemins (2005)
Artic fox
Recognize individuals
DFA
Frommolt et al. (2003)
African wild dog
Recognize individuals
DFA
Hartwig (2005)
Domestic dog
Classify contexts
DFA
Yin and McCowan (2004)
Recognize individuals (breeds)
DFA
Yin and McCowan (2004)
Mudi dog
Classify contexts
Gaussian NB
Molnár et al. (2008)
Recognize individuals
Gaussian NB
Molnár et al. (2008)
ANN artificial neural network, SVM support vector machine, DFA discriminant function analysis, GMM Gaussian mixture model, kNN k-nearest
neighbor classifier, HMM hidden Markov model, NB naive Bayes
vector machines have been employed to automate the
classification of calls of several frog and bird species
(Acevedo et al. 2009). Gaussian mixture models have also
been used for individual animal recognition in birds
(Cheng et al. 2010). Bat species have been acoustically
identified using artificial neural networks (Parsons 2001;
Britzke et al. 2011), discriminant function analysis (Parsons and Jones 2000; Britzke et al. 2011), classification
trees (Adams et al. 2010), k-nearest neighbors (Britzke
et al. 2011) as well as other classifiers (random forests and
support vector machines) whose behavior has been compared (Armitage and Ober 2010). Artificial neural networks
have been used to discriminate between the sounds of
different animals within a group of British insect species
(Orthoptera), including crickets and grasshoppers (Chesmore 2001). Blumstein and Munos (2005) found potentially significant information about identity, age and sex
encoded in yellow-bellied marmots calls using discriminant function analysis. For suricates, discriminant function
analysis was chosen to predict the predator type (mammal,
bird and snake) from the alarm calls (Manser et al. 2002).
Hidden Markov models have been used to analyze African
elephant vocalizations and speaker identification, discrimination of rumbles in different contexts, and oestrous cycle
phase determination from rumbles of female elephants
(Clemins 2005). Moreover, other work has focused on
identifying calls from different animals such as bears,
eagles, elephants, gorillas, lions and wolves, with k-nearest
neighbor classifiers, artificial neural networks and hybrid
methods (Gunasekaran and Revathy 2011).
For canids, research analyzing the acoustic measures of
barks with machine learning methods is limited, see
Table 1. Discriminant functions have been used for individual recognition within a wild population of Arctic foxes
(Frommolt et al. 2003) and African wild dogs (Hartwig
2005). Domestic dog barks have been analyzed again using
discriminant analysis (Yin and McCowan 2004) for classification into context-based subtypes (three different contexts) and in order to identify individual dogs. These two
tasks were further refined in the same paper to categorize
each individual’s barks into separate contexts and identify
the individual barking within each context. A total of 4,672
123
Anim Cogn
barks were recorded from ten dogs of six different breeds,
and 120 variables were extracted from the spectrograms.
More recently, 6,006 barks of 14 Mudi breed individuals
were recorded under six different communicative situations
(Molnár et al. 2008). After processing the spectrograms of
their signals, a genetic programming-based heuristic guided
the construction of new descriptors. The aims were the same
as in Yin and McCowan (2004), although the machine
learning technique was a Gaussian naive Bayes classifier.
In this paper, we extend Molnár et al.’s research in
several ways. As in Molnár et al. (2008), we classify barks
into contexts and identify individual barks. Unlike Molnár
et al., we also investigate whether barks encode information
about dog sex and age. Also, we specify context classification per individual dog and recognize individual bark per
context. Therefore, we have six different classification
problems concerning sex, age, contexts, contexts per individual, individuals and individuals per context. Moreover,
for each of these six problems, a thorough set of four
machine learning models (Gaussian naive Bayes, classification trees, k-nearest neighbors and logistic regression) are
trained from a database of 800 barks corresponding to 8
Mudi dogs in seven behavioral contexts. Their performance
is estimated using cross-validation (K-fold scheme) which
assesses the ability to classify barks that had not been previously encountered. Given an incoming Mudi dog bark,
two models (Gaussian naive Bayes and logistic regression)
output the probability of each class value, whereas the other
two models deterministically provide the predicted class
value. Gaussian naive Bayes assumes normality and independence of the features given the class value. Logistic
regression uses the sigmoid function of a linear combination
of the features as the probability of each class value.
Classification trees hierarchically partition the feature
space. Finally, k-nearest neighbors simply predicts the class
value by majority voting in a feature space neighborhood.
The diversity of these four models is representative of the
available supervised classifiers. Rather than using all the
extracted acoustic measures, we selected relevant features
with two methods, filter and wrapper, for each machine
learning model. Whereas wrapper methods use a predictive
model to score feature subsets, filter methods use a proxy
measure instead of the classification accuracy to score the
selected features.
(Fédération Cynologique Internationale). Initially, we
collected 7,310 barks from 27 individuals. The number of
barks per dog ranged from 8 to 1,696. These barks were
recorded in different number of bouts for each dog. Trying
to minimize the effect of pseudoreplication, we only considered dogs whose initial number of barks was greater than
300. From each of these 8 dogs, 100 barks were randomly
selected using a systematic sampling procedure, thereby
balancing the number of samples coming from each individual. Table 2 contains the characteristics of these selected
800 barks according to sex ratio (male–female 3:5), age
(ranging from 1 to 10 years old), number of bouts for each
dog (with a minimum of 5 and a maximum of 14) and
number of barks per dog in each of the seven contexts. Age
values are grouped into intervals to form a three-valued
class variable: young dogs (1–3 years old), adult dogs
(4–8 years old) and old dogs (more than 8 years old).
Recording and processing of the sound material
Recording contexts
Recordings were made using a Sony TCD-100 DAT tape
recorder and Sony ECM-MS907 microphone on Sony
PDP-65C DAT tapes. During recording of the barks, the
experimenter held the microphone at a distance of 3 to 4
m from the dog. We collected bark recordings in seven
different behavioral contexts. With the exception of two
contexts (Alone and Fight), all recordings were done at
the dog’s residence. Barks of the Fight context were
recorded at dog training schools. The training school dogs
were also taken to a park or other suitable outdoor area to
record the Alone barks. The seven situations are as
follows:
–
–
–
Methods
Subjects
–
Barks recorded from Mudi dogs were used for this study.
The Mudi is a medium-sized Hungarian herding dog breed.
The Mudi breed standard is listed as #238 with the FCI
123
–
Alone (N ¼ 106 recordings): The owner and the
experimenter (male, 23 years old) took the dog to a
park or other outdoor area, where the dog was tied to a
tree or fence by its leash. The owner left the dog and
walked out of the dog’s sight, while the experimenter
remained with the dog and recorded its barks.
Ball (N ¼ 131): The owner held a ball (or one of the
dog’s favorite toys) approximately 1.5 m in front of the
dog.
Fight (N ¼ 131): For dogs to perform in this situation,
the trainer acts as if he intends to attack the dog–owner
dyad. Dogs are expected to bark aggressively and even
bite the trainer’s glove. The owner keeps the dog on a
leash during this exercise.
Food (N ¼ 106): The owner held the dog’s food bowl
approximately 1.5 m in front of the dog.
Play (N ¼ 89): The owner was asked to play a game
with the dog, such as tug-of-war, chasing or wrestling.
Anim Cogn
Table 2 Characteristics of the bark data set with seven context categories: Alone, Ball, Fight, Food, Play, Stranger and Walk
Context
Number
Dog
Sex
Age (years)
Bouts
Alone
Ball
1
Bogyó
Male
1
5
2
Derüs
Female
2
15
3
Fecske
Female
2
10
25
25
4
Guba
Female
5
14
50
50
5
Harmat
Female
4
7
6
Sába
Female
6
7
7
8
Ügyes
Merse
Male
Male
10
7
6
6
Total
–
–
Fight
Food
Play
50
Stranger
Walk
Total
50
100
50
100
25
25
100
50
50
50
100
100
25
25
25
25
17
14
17
14
17
14
17
14
14
17
14
17
14
102
98
106
131
131
106
89
206
31
800
The experimenter recorded the barks emitted during
this interaction.
Stranger (N ¼ 206): The experimenter acted as the
’stranger’ for all the dogs and appeared at the dog
owners’ garden or front door. The experimenter
recorded the barking dog for 2–3 min. The owner
was not in the vicinity (in the garden, or near to the
entrance) during the recording.
Walk (N ¼ 31): We asked the owner to behave as if
he/she was preparing to go for a walk with the dog. For
example, the owner took the dog’s leash in her/his hand
and told the dog ‘We are leaving now.’
Initial processing of the sound material
The recorded material was digitalized with a 16-bit quantization and 44.10 kHz sampling rate using a TerraTec
DMX 6Wre 24/96 sound card. As each recording could
contain at least three or four barks, individual bark sounds
were manually segmented and extracted. This process
resulted in a final collection of 7,310 sound files containing
only a single bark sound. Obviously, these sounds are not
independent from a statistical point of view. As some of the
machine learning methods used in this work assume that
the samples are independent and identically distributed, we
randomly selected non-consecutive barks, alleviating in
this way the pseudoreplication effect. The final data set
contains 800 barks from the initial 7,310 sound files.
Sound analysis
Based on the initial parameter set used in Molnár et al.
(2008), 29 acoustic measures were extracted from the bark
samples with an automated Praat script, see Table 3 and
Fig. 1.
The energy, loudness and the long-term average spectrum (LTAS) are measurements of sound energy, and the
100
LTAS parameters reflect its change over time, whereas the
spectral parameters show the distribution of energy over
the frequency components.
According to the source–filter framework (Fant 1976),
the fundamental frequency is the lowest harmonic component of the source signal that is produced in the larynx by
the movements of the vocal fold. Measurements of the
fundamental show the modulation of this source signal
over time. One voice cycle is the unit of the movements of
the vocal folds. During sound production, the repeated
opening and closing of the vocal folds generates cyclic
pressure changes in the exhaled air, which will be the
sound wave itself. Measurements of the vocal cycles show
the regularities in voice production.
Finally, tonality or harmonics-to-noise ratio gives the
proportion of regular, tonal frequency components over the
noise caused by the irregular movements of the vocal folds,
or the turbulences in the air flow in the vocal tract. These
measurements are capable of describing the quality of the
sound and its change over time.
The process is illustrated in Fig. 2 (top).
Supervised classification
A common machine learning task is pattern recognition
(Duda et al. 2001), in which two different problems are
considered depending on the available information. We
always started from a data set in which each case or
instance (a single bark sound in this paper) is characterized
by features or variables (29 acoustical measures in our
case). In a supervised classification problem, an additional
variable—called the class variable—contains the instance
label (sex, age, context or individual in this paper), and we
look for a model able to predict the label of a new case with
known features. Alternatively, in an unsupervised classification problem or clustering (Jain et al. 1999), the label is
missing and the aim is to form groups or clusters with cases
123
Anim Cogn
Table 3 Twenty-nine acoustic measures extracted from barking
recordings
Name
Description
Variable
Measurements of sound energy
Energy
Amount of energy in the sound
(Pa2 s)
X1
Loudness
Loudness
X10
Ltasm
Mean long-term average spectrum
(ltas)
X23
Ltass
Slope of the ltas
X24
Ltasp
Local peak height between 1,700
and 3,200 in the ltas
X25
Ltasd
Standard deviation of the ltas
X26
Measurements of spectral energy
Banddensity
Density of the spectrum between
2,000 and 4,000 Hz
X2
Centerofgravityfreq
Average frequency in the spectrum
X3
Deviationfreq
Standard deviation of the frequency
in the spectrum
X4
Skewness
Skewness of the spectrum
X5
Kurtosis
Kurtosis of the spectrum
X6
Cmoment
Non-normalized skewness of the
spectrum
X7
Energydiff
Energy difference between 0–2,000
and 2,000–6,000 Hz bands
X8
Densitydiff
Density difference between 0–2,000
and 2,000–6,000 Hz bands
X9
Measurements of the source signal
Pitchm
Mean fundamental frequency (F0)
in Hertz
X11
Pitchmin
Minimum F0
X12
Pitchmax
Maximum F0
X13
Pitchmint
Time point of the minimum F0 (s)
X14
Pitchmaxt
Time point of the maximum F0 (s)
X15
Pitchd
Standard deviation of the F0
X16
Pitchq
Lower interquantile of the F0
X17
Pitchslope
Mean absolute slope of the F0
X18
Pitchslopenojump
Mean slope of the F0 without octave
jump
X19
Measurements of the voice cycles
Ppp
Number of voice cycles
X20
Ppm
Mean number of voice cycles
X21
Ppj
Jitter
X22
Measures of the tonality
Harmmax
Maximum tonality
X27
Harmmean
Mean tonality
X28
Harmdev
Standard deviation of the tonality
X29
(dog barks) that are similar with respect to the features at
hand.
In this paper, we apply supervised classification methods
to automatically learn models from data. These models will
123
be used to separately predict dog sex, dog age, context and
the individual dog from a set of predictor variables capturing the acoustical measures of the dog barks.
In a binary supervised classification problem, there is a
feature vector X 2 Rn whose components, X1 ; . . .; Xn , are
called predictor variables, and there is also a label or class
variable C taking values on f0; 1g. The task is to induce
classifier models from training data, which consists of a set
of N observations DN ¼ fðxð1Þ ; cð1Þ Þ; . . .; ðxðNÞ ; cðNÞ Þg
drawn from the joint probability distribution pðx; cÞ, see
Table 4. In our dog data set, n ¼ 8 acoustical measures and
N ¼ 800 bark sounds. The classification model will be
used to assign labels to new instances, xðNþ1Þ , only characterized by the values of the predictor variables.
To quantify the goodness of a binary classification
model, true positives (TP), true negatives (TN), false positives (FP) and false negatives (FN) are counted over the
test data and placed in a confusion matrix. This confusion
matrix contains in its diagonal the TP and TN observations.
Then, we can define the error rate as jFNjþjFPj
, where N ¼
N
jTPj þ jFPj þ jTNj þ jFNj is the total number of instances,
.
or equivalently, the accuracy as jTPjþjTNj
N
Dog sex classification is binary, XC ¼ {Female, Male},
where there are two possible errors: predict a Male as a
Female dog, and alternatively predict a Female as a Male.
The other classifications are multiclass, where C takes
r [ 2 possible class values. Let XC ¼ f1; 2; . . .; rg denote
this set. Thus, XC ¼ {Young, Adult, Old} for age, XC ¼
{Alone, Ball, Fight, Food, Play, Stranger, Walk} for contexts, and XC ¼ {dog1,...,dog8} for individuals in our case.
The r r-dimensional confusion matrix contains all pairwise counts, mij , the number of cases out of N from the real
class ci classified by the model as cj . The accuracy is given
P
by ri¼1 mii =N.
Accuracy estimation of supervised classification models
An important issue is how to honestly estimate the
(expected) accuracy of a classification model when using
this model for classifying unseen (new) instances. A simple
method is to partition the whole data set into two subsets:
the training subset and the test subset. According to this
training and testing scheme, the classification model is
learned from the training subset, and it is then used in the
test subset for the purpose of estimating its accuracy.
However, the information in the data set is under-used, as
the classification model is learned from a subset of the
original data set.
In this paper, we will use an estimation method called
K-fold cross-validation (Stone 1974). This uses the whole
data set to honestly learn the model. The data set is partitioned into K folds of approximately the same size. Each
Anim Cogn
Fig. 1 Main parameters measured for the acoustic analysis using
Praat functions. The oscillogram shows the actual complex waveform
of a single bark. The amplitude of the waveform shows the intensity
change over time, which is represented here as the intensity profile.
The energy parameter is the overall energy transferred by the sound
over time. Fast Fourier transformation is used to create a sonogram
which shows the frequency spectrum of the bark over time.
Autocorrelation method was applied to extract the fundamental
frequency and its profile depicted as the pitch object. The fundamental frequency is the frequency of opening and closing cycles of the
vocal fold, which is represented by the point process object where
every vertical line represents one vocal cycle. This can be used to
measure the periodicity of the sound and irregularities in sound
production (jitter). The spectrum shows the overall power of each
frequency component. The harmonic-to-noise ratio gives the ratio of
harmonic spectral components (the upper harmonics of the fundamental frequency) over the irregular, noisy components. Finally, the
long-term average spectrum (LTAS) represents the average energy
distribution over the frequency spectrum
fold is left out of the learning process, which is carried out
with the remaining K 1 folds, and used later as a test set.
This process is repeated K times. Thus, every instance is in
a test set exactly once and in a training set K 1 times.
The model accuracy is estimated as the mean of the
accuracies for each of the K test sets. In our experiments,
we will fix the value of K to 10.
This dimensionality reduction by means of an FSS
process has several potential advantages for a supervised
classification model, such as the reduction in the cost of
data acquisition, an improved understanding of the final
classification model, a faster induction of the classification
model and an improvement in classifier accuracy.
FSS can be viewed as a search problem, where each
state in the search space specifies a subset of selectable
features. An exhaustive search of all possible feature subsets, given by 2n , is usually unfeasible in practice because
of the large computational burden, and heuristic search is
usually used.
For a categorization of FSS, see Saeys et al. (2007).
There are two main types of FSS depending on the function
used to measure the goodness of each selected subset. In
the wrapper approach to the FSS, the accuracy reported by
Feature subset selection
The feature subset selection (FSS) problem (Liu and Motoda 1998) refers to the question of whether all the n predictor features are really useful for classifying the instances
with a given model. The FSS problem can be formulated as
follows: Given a set of candidate features, select the best
subset under some classification learning method.
123
Anim Cogn
Fig. 2 Diagram of the study: data preprocessing (top) and questions to be answered by machine learning models (bottom)
Table 4 Raw data in a supervised classification problem: N denotes
the number of labeled observations, each of them characterized by n
predictor variables, X1 ; . . .; Xn and the class variable C
...
Xn
C
ð1Þ
...
xð1Þ
n
cð1Þ
ð2Þ
xð2Þ
n
cð2Þ
X1
(xð1Þ ; cð1Þ )
x1
(xð2Þ ; cð2Þ )
x1
...
ðNÞ
x1
ðNþ1Þ
x1
...
xðNÞ
n
cðNÞ
...
xðNþ1Þ
n
?
...
(xðNÞ ; cðNÞ )
xðNþ1Þ
...
...
xðNþ1Þ denotes the new observation to be classified by the supervised
classification model
a classifier guides the search for a good subset of features.
We have used a greedy stepwise search in our experiments, i.e., one that progresses forward from the empty
set selecting at each step the best option among adding a
variable not yet included within the model and deleting a
variable from the current model. The search is halted
when neither of these options improves model accuracy.
When the learning algorithm is not used in the evaluation
function, the goodness of a feature subset can be assessed
using only intrinsic data properties, such as an information
theory based evaluation function. This is the filter
approach to the FSS problem. In this paper, we apply both
123
wrapper and filter approaches to the FSS problem. For the
second type, a multivariate filter based on mutual information, called correlation feature selection, is used (Hall
1999). This tries both to minimize redundancy between
selected features and maximize correlation with the class
variable.
Supervised classification methods
Given an instance x, supervised classification builds a
function c that assigns to x a class label in XC ¼ f1; . . .; rg.
We provide a short description of each supervised classification method used.
Naive Bayes (Minsky 1961) is the simplest Bayesian
classifier. A Bayesian classifier assigns the most probable a
posteriori class to a given instance x, i.e., it yields the c
value of C that maximizes the posterior probability pðcjxÞ.
Using the Bayes’ theorem, this is equivalent to maximizing
pðcÞpðxjcÞ. The naive Bayes is built upon the assumption
of conditional independence of the predictive variables
given the class. Computationally, this means that pðxjcÞ in
the previous product is easily obtained as the product of all
factors pðxj jcÞ; j ¼ 1; . . .; n, each associated with one variable. The Gaussian naive Bayes classifier applies for
continuous variables Xj following a Gaussian distribution
fj . Therefore, this model computes c such that
Anim Cogn
max pðcÞ
c2XC
n
Y
fj ðxj jcÞ:
ð1Þ
j¼1
In a classification tree (Quinlan 1993), the learned function
c is represented by a decision tree. Each (non-leaf) node
specifies a value test of some variable of the instance. Each
descendant branch corresponds to one of the possible values for this variable. Each leaf node provides the class label
given the values of the variables jointly represented by the
path from the root to that leaf. Unseen instances are classified by sorting down the tree from the root to some leaf
node testing the variable specified at each node. A classification tree is learned in a top-down manner (starting with
the root node) by progressively splitting the training data
set into smaller and smaller subsets based on variable value
tests. This process is repeated on each derived subset in a
recursive manner called recursive partitioning of the space
representing the predictive variables. Key decisions are
how to select which variable to test at each node in the tree,
and how deep the tree should be, i.e., whether to stop
splitting or select another variable and grow the tree further. These decisions make the differences between algorithms. The C4.5 algorithm used in this paper chooses
variables by maximizing the gain ratio, which is the ratio of
the information gain of Xj and C and the entropy of Xj ,
which are both concepts used in information theory. The
algorithm incorporates post-pruning rules to avoid the tree
becoming too deep thereby escaping from the training data
overfitting, i.e., its failure to work well with new unseen
instances.
The k-nearest neighbor classifier (Fix and Hodges 1951)
is a nonparametric method that assigns to a given instance
x the class label most frequently found among its k nearest
instances; that is, the predicted class is decided by examining the labels of the k nearest neighbors and voting. A
common distance used for obtaining the k nearest neighbors for a continuous variable x is the Euclidean distance.
This classifier is a type of lazy learning where the function
is only approximated locally and all computation is
deferred until classification. In our experiments, we will fix
k ¼ 1.
Logistic regression (Le Cessie and van Houwelingen
1992), like naive Bayes, produces a posterior probability
pðcjxÞ for a given instance x. For binary classification, the
model assumes that it is a transformation of a linear
combination of the input variables, given by
pðC ¼ 1jxÞ ¼ 1=½1 þ eðb0 þb1 x1 þþbn xn Þ ;
where b0 ; b1 ; . . .; bn are model parameters estimated from
data by maximum likelihood. If Lðb0 ; . . .; bn Þ denotes the
log-likelihood function of the data under this model, the
problem is to find bs that maximize this function. The ridge
logistic regression used in this paper adds a penalization
term to L, and the problem is then to maximize the funcP
tion Lðb0 ; . . .; bn Þ k nj¼1 b2j , for bs where k [ 0 controls the amount of penalization. This penalty forces the
parameters to shrink to zero achieving a reduction in the
variance of the parameter estimates with an overall
increased accuracy. For multiclass classification, the posterior probability of c 6¼ r is given by
ðcÞ
ðcÞ
ðcÞ
eðb0 þb1 x1 þþbn xn Þ
pðcjxÞ ¼
Pr1 ðbðlÞ þbðlÞ x1 þþbðlÞ xn Þ ;
n
1 þ l¼1 e 0 1
l ¼ 1; . . .; r 1
ð2Þ
and hence, pðrjxÞ is derived from the others since they all
sum to one. Note that in this multiclass case, we need a set
ðlÞ
ðlÞ
of n þ 1 parameters fb0 ; b1 ; . . .; bðlÞ
n g for each l value,
l ¼ 1; . . .; r 1; that is, a total of ðn þ 1Þðr 1Þ
parameters.
All the results were calculated using WEKA software
(Hall et al. 2009).
Results
The six problems we will deal with are illustrated in Fig. 2
(bottom).
Sex
The k-nearest neighbor classifier produced the best results,
with an accuracy of 85.13 %, with a wrapper feature
selection (in bold), see Table 5. This model contains 12
predictor variables, see Table 16. The groups that record
spectral energy and source signal variables are under-represented, according to the categorization of acoustic measures provided in Table 3.
For the female barks, the misclassification rate is 9.40 %
(47 false males out of 500 real females), and it is higher for
males, 24.00 % (72 false females from a total of 300 real
males).
Table 6 shows the accuracies per dog of the k-nearest
neighbor model with 12 predictors. The model accuracy
when predicting the five female dogs is around 90 %, with
the worst predictions for dog3 and dog4 (87.00 %), and the
best for dog5 (97.00 %). The three male dogs are predicted
with accuracies ranging from 73.00 % for dog1 to 79.41 %
for dog7.
Supplementary Material contains the specifications of
the best models for the prediction of the dog sex. For naive
Bayes, the univariate conditional Gaussian densities for
each predictor variable are shown. The structure of the
classification tree model is also presented, as well as the
123
Anim Cogn
Table 5 Sex prediction
Table 7 Age prediction
All
Filter
Wrapper
Naive Bayes
68.50 %
65.63 %
71.88 %
Classification tree
70.88 %
69.13 %
74.13 %
85.13 %
k-Nearest neighbors
78.63 %
79.13 %
80.25 %
78.63 %
Logistic regression
75.63 %
73.88 %
76.00 %
All
Filter
Wrapper
Naive Bayes
71.00 %
71.13 %
77.13 %
Classification tree
78.13 %
72.75 %
81.50 %
k-Nearest neighbors
82.00 %
64.25 %
Logistic regression
76.88 %
70.50 %
Accuracies of the twelve models: three selection feature methods for
each of the four supervised classifiers
Real class
Predicted class
Young
Table 6 Sex prediction per dog
Accuracies of the best model in
Table 5 for each of the eight
dogs. The overall accuracy of
this model over the eight dogs is
85.13 %
Male
76.00 %
Female
90.60 %
Dog1
73.00 %
–
Dog2
–
90.00 %
Dog3
–
87.00 %
Dog4
–
87.00 %
Dog5
–
97.00 %
Dog6
–
92.00 %
Dog7
79.41 %
–
Dog8
75.51 %
–
coefficients of the logistic regression model. For the knearest neighbor classifier, the data set constitutes the
model and therefore it is not shown.
Age
Table 7 (left) shows the age results. As for the sex prediction problem, k-nearest neighbors with a wrapper feature selection produced the best accuracy 80.25 %. The 15
selected variables in this model mainly contain measurements of spectral energy, sound energy and voice cycles.
For this problem, the wrapper strategy outperformed the
other strategies in the four supervised classification
methods.
The confusion matrix in Table 7 (right) of the best
model shows that a Young dog is classified as Old in only
2.67 % of cases (8 out of 300), while old dogs are misclassified as Young in 6.86 % of cases (7 out of 102). The
error rates classifying Young, Adult and Old dogs are
21.00, 17.59 and 24.51 %, respectively. These figures
suggest that it is easier to get it wrong when classifying
Young and Old dogs.
Table 8 contains the accuracies per dog of the best
model. This model provides a 79.00 % of accuracy when
predicting Young dogs. This percentage is very similar for
each of the three young dogs (dog1, dog2 and dog3).
However, for the four adult dogs the model shows a wide
range of accuracies, varying from 66.00 % (dog6) to
90.00 % (dog5). Dog7, that is the only old dog, is classified
with an accuracy of 75.49 %.
123
Adult
Old
8
Young
237
55
Adult
61
328
9
7
18
77
Old
Accuracies of the twelve models: three selection feature methods for
each of the four supervised classifiers (top table). Confusion matrix of
the best model: k-nearest neighbors wrapper (bottom table)
Table 8 Age prediction per dog
Young 79.00 %
Adult 82.41 %
Old 75.49 %
Dog1
84.00 %
–
–
Dog2
74.00 %
–
–
Dog3
79.00 %
–
–
Dog4
–
85.00 %
–
Dog5
–
90.00 %
–
Dog6
–
66.00 %
–
Dog7
Dog8
–
–
–
88.77 %
75.49 %
–
Accuracies of the best model in Table 7 for each of the eight dogs.
The overall accuracy of this model over the eight dogs is 80.25 %
Supplementary Material contains the specifications of
the best models for the prediction of the dog age.
Context
A single model for all dogs. Table 9 (left) shows the results
of a single model learned from the 800 barks to discriminate among the 7 contexts: Alone, Ball, Fight, Food, Play,
Stranger and Walk.
k-nearest neighbor classifier and wrapper selection is
once more the best-performing model with an accuracy of
55.50 %. The variables selected by this model correspond
mainly to spectral energy and voice cycle measurements.
Note that now we have a more difficult problem with more
class values to be predicted (7 contexts) and consequently
the estimated accuracy is expected to be lower.
From Table 9 (right), we can compute the contexts with
the highest and lowest true positive rates that correspond to
Fight (0.76) and Walk (0.35), respectively. The Ball
Anim Cogn
context is often misclassified as Food and vice versa. The
same holds for the Walk and Play pair. This is quite reasonable since both pairs define quite similar underlying
concepts. Many barks under Fight or Alone situations are
misclassified as Stranger. However, the Stranger context is
usually confused with the Ball and Food context.
Table 10 contains the accuracies per dog of the best
model. This model provides 43.40 % accuracy when
predicting the Alone context, with extreme prediction
accuracies for dog7 (52.94 %) and dog8 (14.29 %). The
Ball context achieves 48.85 % accuracy, having dog7 and
dog8 the worst (29.41 %) and best (64.29 %) predictions,
respectively. These two dogs also present the worst and
best predictions for the Food context. The model shows
better accuracies for the Fight and Stranger contexts. In
the Fight context, the 98.00 % of success for dog5 is
noteworthy, whereas the worst behavior in the Stranger
context is for dog7 (35.29 %). The Play and Walk contexts show highly variable accuracies for the different
dogs.
Supplementary Material contains the specifications of
the best models for the prediction of the dog context.
A model per dog. More refined dog-specific models are
built here. By selecting instances from the same dog, the
corresponding model will identify the context for that dog.
A total of 96 models (8 dogs 12 models per dog) have
been considered, where only the performance of the best
model is shown in Table 11.
Naive Bayes was the best model 3 times, k-nearest
neighbors 4 times, and logistic regression in 2 cases.
Regarding the feature subset selection methods, wrapper
reports the best results for all 8 dogs.
Table 11 shows that accuracies decrease in proportion to
the increase in the number of contexts. With two contexts,
accuracies fall in the interval [78, 100 %]. The accuracies
for the two dogs with four contexts are 74 and 73 %.
Increasing the number of contexts to six and seven, the
accuracies are 59.80 and 66.98 %, respectively.
Figure 3 displays, for the best models in Table 11, the
mean number of selected variables by the five types of
acoustic variables. Spectral energy and voice cycle measurements were the two groups with more often selected
(in relative terms) variables regardless of the number of
barks.
From the previous table, we select some models for the
sake of illustration. Figure 4 shows the naive Bayes
model which performed best for dog5, with only two
observed contexts, Fight and Strange (see the first row in
Table 11). The model is built with only five variables,
Deviationfreq, Pitchmax, Pitchmaxt, Pitchd and Ppp
selected by the wrapper approach. The missing arcs
between predictor variables and the arcs from the class to
the predictor variables encode the assumption of conditional independence underlying naive Bayes. Figure 4
also shows the parameters, pðcÞ and the mean and standard deviation of the Gaussian distributions fj ðxj jcÞ in
Eq. (1).
Table 10 Context prediction
per dog
Table 9 Context prediction
All
Filter
Wrapper
Naive Bayes
41.63 %
42.63 %
47.88 %
Classification tree
44.00 %
44.63 %
44.13 %
k-Nearest neighbors
50.88 %
50.75 %
55.50 %
Logistic regression
49.75 %
47.50 %
50.13 %
Real class
Predicted class
Alone
Ball
Alone
46
15
Ball
11
Fight
8
Food
Play
Stranger
Walk
Fight
Food
Play
Stranger
Walk
7
17
6
14
1
64
5
22
5
23
1
4
100
3
4
11
1
7
20
2
55
3
15
4
8
8
2
10
44
11
6
12
24
5
26
13
124
2
0
3
4
5
6
2
11
Accuracies of the twelve models: three selection feature methods for
each of the four supervised classifiers (top table). Confusion matrix of
the best model: k-nearest neighbors wrapper (bottom table)
Accuracies of the best model in
Table 9 for each of the eight
dogs. The overall accuracy of
this model over the eight dogs is
55.50 %
Alone
43.40 %
Ball
48.85 %
Fight
76.34 %
Food
51.89 %
Play
49.44 %
Stranger
60.19 %
Walk
35.48 %
Dog1
–
–
–
–
76.00 %
68.00 %
–
Dog2
–
–
–
64.00 %
–
54.00 %
–
Dog3
52.00 %
44.00 %
56.00 %
–
–
60.00 %
–
Dog4
44.00 %
60.00 %
–
–
–
–
–
Dog5
–
–
98.00 %
–
–
70.00 %
–
Dog6
–
36.00 %
76.00 %
44.00 %
12.00 %
–
–
Dog7
52.94 %
29.41 %
52.94 %
17.65 %
–
35.29 %
41.18 %
Dog8
14.29 %
64.29 %
57.14 %
64.29 %
21.43 %
50.00 %
14.29 %
123
Anim Cogn
Table 11 Context discrimination: A model per dog
Dog
Model
Accuracy
Context
Dog5
Naive Bayes wrapper
k-Nearest neighbors wrapper
100.00 %
Fight Stranger
Dog1
k-Nearest neighbors wrapper
97.00 %
Play Stranger
Dog2
Logistic regression wrapper
86.00 %
Food Stranger
Dog4
Naive Bayes wrapper
78.00 %
Alone Ball
Dog3
k-Nearest neighbors wrapper
74.00 %
Alone Ball Fight Stranger
Dog6
Logistic regression wrapper
73.00 %
Ball Fight Food Play
Dog7
Naive Bayes wrapper
59.80 %
Alone Ball Fight Food Stranger Walk
Dog8
k-Nearest neighbors wrapper
66.98 %
Alone Ball Fight Food Play Stranger Walk
Summary of the best models, accuracies and corresponding contexts for each dog. Dogs are organized by number of contexts and then by model
accuracy
2.5
2.0
1.5
1.0
0.5
ðcÞ
bj for these variables would be used as in Eq. (2) to
compute the posterior probability that yields the predicted
class.
Individual
0.0
Mean number of selected variables
3.0
although its nearest neighbor bark should be computed in
the 5-D space, also including variables Deviationfreq and
Harmmean.
Table 12 includes the details of the logistic regression
model which performed best for dog2, with two observed
contexts, Food and Stranger (see the second row in
Table 11). This model is built from the five predictor
variables in the first column. The regression coefficients
Sound
Spectral
Signal
Voice
Tonality
Groups of acoustic measures
Fig. 3 Mean number of variables (Y-axis) selected by the best models
per dog when predicting contexts (listed in Table 11), for each of the
five groups of acoustic measures (X-axis): sound energy, spectral
energy, source signal, voice cycles and tonality. Each of these groups
of acoustic measures contain 6, 8, 9, 3 and 4 variables, respectively
Figure 5 displays the classification tree model which
performed second best for dog1, with two observed contexts, Play and Stranger (see the second row in Table 11).
Note that three variables are required: Energydiff, Harmmean and Ppj. Thus, if for a given bark, Energydiff = 10,
Harmmean = 15 and Ppj = 0.05, then the dog is classified as
barking at a stranger.
Figure 6 shows the 100 barks recorded for dog1, represented as a point in the 3-D space of three of the five
variables selected by the best model, a k-nearest neighbors
wrapper. Barks in the Play context are colored blue (dark),
whereas Stranger is shaded red (light). A new bark (an
asterisk in the figure) would be classified as the context of
its nearest neighbor bark, i.e., Play in this 3-D space,
123
A single model for all contexts. Table 13 shows the results
of a single model learned from the 800 barks for discriminating among the 8 dogs.
k-nearest neighbors wrapper is the best model, as in the
three previous classification problems, with an extremely
high accuracy, 67.63 %, in an 8 multi-class problem. Thus,
feature subset selection methods have been proved to
produce improvements in model performance.
The true positive rate for each of the classes can be
computed from Table 14. Dogs numbers 8, 5 and 7 have
high true positive rates: 0.77, 0.75 and 0.74, respectively.
In contrast, dogs number 6 and 3 have the lowest true
positive rates 0.51 and 0.58, respectively.
A model per context. More refined context-specific
models are built here. By selecting bark sounds from the
same context, the corresponding model will classify the
individual dog for that context. Thus, a total number of 7
contexts (and their corresponding 12 7 models) have
been considered, where the accuracy of the best model for
each context is shown (see Table 15).
Note that the model accuracies for identifying dogs have
increased to an 80–100 % range compared with the
67.63 % achieved by the global model learned from a
Anim Cogn
Fig. 4 Example of a naive
Bayes wrapper model. It
corresponds to the best model
for context classification in
dog5
Table 12 Example of parameter values of a logistic regression model
ðFoodÞ
Variable Xj
bj
Kurtosis
-0.0008
Pitchd
-0.0002
Pitchslope
It corresponds to the best model
for context classification in
dog2
0.0001
Ppp
-0.0143
Ppm
-7,424.9241
Intercept (b0 )
31.5997
Table 13 Individual prediction
Fig. 5 Example of a classification tree wrapper model. It corresponds
to the second best model for context classification in dog1
All
Filter
Wrapper
Naive Bayes
54.50 %
55.63 %
63.00 %
Classification tree
53.13 %
51.37 %
56.37 %
k-Nearest neighbors
Logistic regression
63.87 %
63.00 %
58.62 %
61.75 %
67.63 %
65.75 %
25
Accuracies of the twelve models: three selection feature methods for
each of the four supervised classifiers
20
cmoment 15
Table 14 Confusion matrix for the best model, k-nearest neighbors
wrapper, identifying individual dogs
10
Dog
5
Predicted class
1
−5
−10
−15
energydiff −20
−25
−30
Real class
1200
1000
800
600
play
pitchq
stranger
Fig. 6 Example of a k-nearest neighbors wrapper model. It corresponds to the best model for context classification in dog1 (Cmoment
scale is divided by 109 ). Classification of a hypothetical bark
(asterisk)
database with all the contexts. We now have fewer dogs to
be identified, from 2 dogs for the Walk context to 5 dogs
for Ball and Fight contexts, whereas the global model had
2
3
4
5
6
7
8
1
68
10
8
5
0
5
3
2
6
71
5
1
5
8
2
1
2
3
9
8
58
6
2
14
1
2
4
7
3
4
67
2
9
2
6
5
1
6
3
2
75
7
6
0
6
5
12
11
6
6
51
9
0
7
2
3
2
5
5
10
74
1
8
1
4
1
8
1
4
2
77
the harder problem of identifying 8 dogs. Although the
problem is easier because there are fewer class variable
values, barking is expected to be homogeneous in a fixed
context, which complicates correct dog identification.
123
Anim Cogn
Table 15 Summary of the results of classifying individuals by
context
Context
No. barks
No. dogs
Alone
106
4
94.34 %
Ball
131
5
Fight
131
5
Accuracy
Table 16 Predictor variables of sex, age, context and individual
classification problems from the best model, k-nearest neighbors
wrapper
Var
Name
Sex
Age
Context
Individual
80.92 %
X1
Energy
x
x
x
x
88.55 %
X10
Loudness
x
x
Ltasm
Ltass
x
x
x
x
Food
106
4
87.74 %
Play
89
3
97.75 %
X23
X24
206
5
80.58 %
X25
Ltasp
x
x
100.00 %
X26
Ltasd
x
x
X2
Banddensity
X3
Centerofgravityfreq
x
X4
Deviationfreq
x
X5
Skewness
x
X6
Kurtosis
X7
Stranger
Walk
31
2
k-nearest neighbors performed best for Alone, Ball, Play and Walk
contexts, naive Bayes for Fight and Walk, logistic regression for
Stranger and Walk, and classification trees for Food context. All best
models corresponded to a wrapper feature subset selection strategy
Predictor variables of sex, age, context and individual
The number of selected variables in the best models (see
Table 16) represents about 50 % of the 29 initial variables.
These numbers were 12 for sex, 15 for age, 16 for context
and 18 for individual. It is remarkable that some variables,
like Ltasm, Ltass, Pitchmint, Pitchslopenojump and
Harmmax were never chosen. On the other hand, the following six variables occur in all four models: Energy,
Ltasp, Deviationfreq, Skewness, Pitchq and Harmmean.
Harmdev appears to be specific for determining dog sex,
since it was not selected in the rest of the problems. This
also applies to Pitchd, only selected for discriminating dog
age and to Pitchmaxt for individual determination.
Considering the blockwise organization of predictor
variables in Table 3, sound energy (first block), source
signal (third block) and tonality (fifth block) measurements
are sparsely selected compared to a denser selection in the
remaining blocks.
x
x
x
x
x
x
x
x
x
x
x
x
Cmoment
x
x
x
X8
Energydiff
x
x
x
X9
Densitydiff
x
X11
Pitchm
X12
Pitchmin
X13
Pitchmax
X14
X15
Pitchmint
Pitchmaxt
X16
Pitchd
X17
Pitchq
X18
Pitchslope
X19
Pitchslopenojump
X20
Ppp
X21
Ppm
X22
Ppj
X27
Harmmax
X28
Harmmean
x
X29
Harmdev
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
Discussion
The accuracies of these four models are 85.13 % for sex classification
(Table 5), 80.25 % for age prediction (Table 7), 55.50 % for context
categorization (Table 9) and 67.63 % for individual recognition
(Table 13)
This work has empirically demonstrated the usefulness of
supervised classification machine learning methods for
inferring some characteristics of dogs from the acoustic
measurements given by their barks. From the four classification methods considered, k-nearest neighbors outperformed naive Bayes, classification trees and logistic
regression. Also, the wrapper feature subset selection
method provided significant improvements over a filter
selection or no-selection (all variables are kept).
A solution for two prediction problems, sex and age,
never previously considered in the literature has been
presented. The best of the 12 resulting models in this study
was able to predict dog sex in 85.13 % of the cases. The
age of the dog, categorized as Young, Adult and Old, was
inferred correctly in 80.25 % of the cases. An issue to be
considered as future work is the prediction of age as a
continuous variable, using a kind of regression task.
Determining the context of the dog bark, with seven
possible situations, is a more difficult problem than classification by sex and age. However, it was successfully
solved for 55.50 % of the bark cases. This is an improvement on the results presented in Molnár et al. (2008), where
for six possible contexts the best model yielded a 43 %
success rate. With an accuracy rate of 63 % for classifying
three possible contexts, our results are similar to the findings reported by Yin and McCowan (2004). In addition, a
model for each of the eight dogs with two or more different
123
Anim Cogn
contexts was induced from the barks associated with this
specific dog. Thus, a total of 12 8 models have been
considered. For almost all dogs, the k-nearest neighbor
model was the most successful, although naive Bayes,
logistic regression and classification tree models provided
the best accuracy results for some dogs. As a tendency, the
wrapper feature subset selection strategy provided the best
results. Model accuracy ranges from 59.80 to 100 %.
The individual identification, a hard classification
problem with eight possible categories, produced up to
67.63 % accuracy in the best model. This result is extremely good when compared to the 52 % reported in Molnár
et al. (2008) for 14 dogs, and the 40 % achieved by Yin and
McCowan (2004) for a 10-dog problem. When the dog
identification is performed within each context, the accuracies of the best models are in the interval [80.58 %,
100 %].
Recent ethological research on dog barking revealed
several features of the most characteristic acoustic communication type of dogs which proved that barks serve as
a complex source of information for listeners (Yin and
McCowan 2004; Pongrácz et al. 2005, 2006). In experiments where human participants evaluated the pre-recorded dog barks, both the context and the possible inner
state of the signaling animals were classified with substantial success rates. However, the role of dog barks in
dog–dog communication remained (and still remains)
somewhat obscure, as there is a shortage of convincing
field data for the usage of barks during intraspecific
communication of dogs, though see Pongrácz et al. (2014)
for some positive evidence. The present study provides an
alternative approach for discovering the potential information content encoded in dog barks. If one can prove that
dog barks carry consistent cues encoding such features of
the caller such as its sex, age or identity, this can prove
indirectly that barks can serve as relevant sources of
information to receivers that are able to decipher these
types of information.
Previously, it was known that dogs can differentiate
between individuals and contexts if they hear barks of other
dogs in experiments based on the habituation–dishabituation paradigm (Maros et al. 2008; Molnár et al. 2009). Our
new results provide some possible details of how such a
capacity for recognition might work. If dogs are sensitive
to the sex-, age- and identity-specific details of barks, this
can serve as an acoustic basis for the cognitive task of
discriminating between or recognition of individuals.
Although in dogs sex-related information is mostly
(thought to be) transferred via chemical compounds
(Goodwin et al. 1979), theoretically it would be adaptive if
a dog could survey the gender of the other dogs living
nearby (or farther) on the basis of hearing their barks as
well. Deciphering the age of an individual based on their
vocalizations would be also beneficial in a highly social
species, where age can be relevant in determining social
rank, reproductive status or fighting potential (Mech 1999).
Recognition of the context of barks was the least successful task for our supervised learning methods. Although
present methods exceeded the accuracy of both the previously employed machine learning approach (Molnár et al.
2008) and the adult human listeners’ success rate (Pongrácz et al. 2005), this accuracy still lags behind the other
variables analyzed in this study. It is also true that human
listeners perform almost as successfully when recognizing
the context as the computerized models. The reason behind
this result may be that the individual variability of dog
barks can be considerable especially in particular contexts
(such as before the walk, or asking for a toy/food). Another
reason for the relatively low success rate of context recognition may be that while the human listeners received
short bark sequences, the computer worked with individual
bark sounds. Therefore, the inter-bark interval served as an
additional source of information for the humans (Pongrácz
et al. 2005, 2006), while this parameter was not involved in
the computerized analysis. For humans at least, the interbark interval also seemed to be an important source of
information when discriminating between individual dogs,
as their performance improved with the length of bark
sequences they received (Molnár et al. 2006).
Supervised classification machine learning methods do
not only provide indirect proof about the rich and biologically relevant information content of dog barks, but they
also offer a promising tool for applied research, too. For
example, evaluating dog behavior has great importance for
various organizations, as well as professionals and dog
enthusiasts. Recognizing unnecessarily aggressive dogs can
be a challenge for the personnel of dog shelters as well as
for correspondents of breed clubs and for the experts of
legal bodies (Netto and Planta 1997; Serpell and Hsu
2001). Similarly, diagnosing particular behavioral abnormalities that can cause serious welfare issues for dogs, such
as separation anxiety, can present a difficult task when the
goal is to tell apart ’everyday’ and chronic stress reactions
in a dog (Overall et al. 2001). Behavioral evaluation usually does not cover the qualitative analysis of vocalizations
in these cases. However, this could be addressed if a reliable and easy to use acoustic analytic software could serve
as an aid for behavioral professionals. With such a method,
following a rigorous validating protocol, acoustic features
indicative of high levels of aggression, fear, distress, etc.
could be recognized in the subjects’ vocalizations.
The limitations of the supervised classification models
presented in this paper concern the standard problems with
the sample representativeness and the assumptions upon
which the models rely. On the other hand, the generality of
the four methods makes them directly applicable to other
123
Anim Cogn
species. In addition, all the dogs in this study were of the
same breed, so our classifiers do not take any advantage of
the different patterns expected from the diversity of breeds.
An interesting problem for the near future would be to
see whether these methods would work for other breeds or
for a mixed breed group. Also, simultaneously classifying
the four dog features, sex, age, context and individual,
might be of interest. This issue falls into a category of a
new problem type called multi-dimensional classification
problems (Bielza et al. 2011; Borchani et al. 2012; Sucar
et al. 2014), where the dependence between the four class
variables is relevant.
Acknowledgments This work has been partially supported by the
Spanish Ministry of Economy and Competitiveness, projects Cajal
Blue Brain (C080020-09; the Spanish partner of the Blue Brain
Project initiative from EPFL) and TIN2013-41592-P, by the János
Bolyai Research Scholarship from the Hungarian Academy of Sciences, and co-financed by a grant from OTKA K82020. The authors
are thankful to Celeste R. Pongrácz for the English proofreading of
this manuscript and to Nikolett Czinege for help in preparing the bark
databases.
References
Acevedo M, Corrada-Bravo C, Corrada-Bravo H, Villanueva-Rivera
L, Aide T (2009) Automated classification of bird and amphibian
calls using machine learning: a comparison of methods. Ecol
Inform 4(4):206–214
Adachi I, Kuwahata H, Fujita K (2007) Dogs recall their owner’s face
upon hearing the owner’s voice. Anim Cogn 10:17–21
Adams M, Law B, Gibson M (2010) Reliable automation of bat call
identification for Eastern New South Wales, Australia, using
classification trees and AnaScheme software. Acta Chiropterol
12(1):231–245
Armitage D, Ober H (2010) A comparison of supervised learning
techniques in the classification of bat echolocation calls. Ecol
Inform 5(6):465–473
Au W, Andersen L, Roitblat ARH, Nachtigall P (1995) Neural
network modeling of a dolphin’s sonar discrimination capabilities. J Acoust Soc Am 98:43–50
Bálint A, Faragó T, Dóka A, Miklósi A, Pongrácz P (2013) ‘‘Beware,
I am big and non-dangerous!’’—playfully growling dogs are
perceived larger than their actual size by their canine audience.
Appl Anim Behav Sci 148:128–137
Bielza C, Li G, Larrañaga P (2011) Multi-dimensional classification
with Bayesian networks. Int J Approx Reason 52:705–727
Blumstein D, Munos O (2005) Individual, age and sex-specific
information is contained in yellow-bellied marmot alarm calls.
Anim Behav 69(2):353–361
Borchani H, Bielza C, Martı́nez-Martı́n P, Larrañaga P (2012)
Markov blanket-based approach for learning multi-dimensional
Bayesian network classifiers: an application to predict the
European Quality of Life-5 Dimensions (EQ-5D) from the
39-item Parkinson’s Disease Questionnaire (PDQ-39). J Biomed
Inform 45:1175–1184
Britzke E, Duchamp J, Murray K, Swihart R, Robbins L (2011)
Acoustic identification of bats in the Eastern United States: a
comparison of parametric and nonparametric methods. J Wildl
Manage 75(3):660–667
123
Charrier I, Aubin T, Mathevon N (2010) Mother-calf vocal communication in Atlantic walrus: a first field experimental study. Anim
Cogn 13(3):471–482
Cheng J, Sun Y, Ji L (2010) A call-independent and automatic
acoustic system for the individual recognition of animals: a
novel model using four passerines. Pattern Recogn
43(11):3846–3852
Chesmore E (2001) Application of time domain signal coding and
artificial neural networks to passive acoustical identification of
animals. Appl Acoust 62(12):1359–1374
Clemins P (2005) Automatic Classification of Animal Vocalizations.
PhD thesis, Marquete University
Cohen J, Fox M (1976) Vocalizations in wild canids and possible
effects of domestication. Behav Process 1:77–92
Coppinger R, Feinstein M (1991) ‘‘Hark! Hark! the dogs barkldots’’
and bark and hark. Smithonian 21:119–128
Druzhkova A, Thalmann O, Trifonov V, Leonard J, Vorobieva N,
Ovodov N, ASGraphodatsky, Wayne R (2013) Ancient DNA
analysis affirms the canid from Altai as a primitive dog. PLoS
ONE 8(e57):754
Duda R, Hart P, Stork D (2001) Pattern classification. Wiley, New York
Fant G (1976) Acoustic theory of speech production. Mouton De
Gruyter.
Faragó T, Pongrácz P, Miklósi A, Huber L, Virányi Z, Range F
(2010a) Dogs’ expectation about signalers’ body size by virtue
of their growls. PLoS ONE 5(12):e15,175
Faragó T, Pongrácz P, Range F, Virányi Z, Miklósi A (2010b) The
bone is mine’: affective and referential aspects of dog growls.
Anim Behav 79(4):917–925
Feddersen-Petersen DU (2000) Vocalization of European wolves
(Canis lupus lupus l.) and various dog breeds (Canis lupus f.
fam.). Arch Tierz Dummerstorf 43(4):387–397
Fix E, Hodges JL (1951) Discriminatory analysis. Nonparametric
discrimination: consistency properties. USAF Sch Aviat Med
4:261–279
Frommolt KH, Goltsman M, MacDonald D (2003) Barking foxes,
Alopex lagopus: field experiments in individual recognition in a
territorial mammal. Anim Behav 65:509–518
Goodwin M, Gooding KM, Regnier F (1979) Sex pheromone in the
dog. Science 203:559–561
Gunasekaran S, Revathy K (2011) Automatic recognition and
retrieval of wild animal vocalizations. Int J Comput Theor Eng
3(1):136–140
Hall M (1999) Correlation-based feature selection for machine
learning. PhD thesis, Department of Computer Science, University of Waikato, UK
Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten I
(2009) The WEKA data mining software: an update. SIGKDD
Explor 11(1):10–18
Hartwig S (2005) Individual acoustic identification as a non-invasive
conservation tool: an approach to the conservation of the African
wild dog Lycaon pictus (Temminck, 1820). Bioacoustics
15:35–50
Hecht J, Miklósi A, Gácsi M (2012) Behavioral assessment and owner
perceptions of behaviors associated with guilt in dogs. Appl
Anim Behav Sci 139:134–142
Hunag C, Yang Y, Yang D, Chen Y (2009) Frog classification using
machine learning techniques. Expert Syst Appl 36(2):3737–3743
Jain A, Murty MN, Flynn P (1999) Data clustering: a review. ACM
Comput Surv 31(3):264–323
Le Cessie S, van Houwelingen JC (1992) Ridge estimators in logistic
regression. Appl Stat 41(1):191–201
Liu H, Motoda H (1998) Feature selection for knowledge discovery
and data mining. Kluwer Academic, Dordrecht
Lord K, Feinstein M, Coppinger R (2000) Barking and mobbing.
Behav Process 81:358–368
Anim Cogn
Manser M, Seyfarth R, Cheney D (2002) Suricate alarm calls signal
predator class and urgency. Trends Cogn Sci 6(2):55–57
Maros K, Pongrácz P, Bárdos G, Molnár C, Faragó T, Miklósi A
(2008) Dogs can discriminate barks from different situations.
Appl Anim Behav Sci 114:159–167
Mazzini F, Townsend SW, Virányi Z, Range F (2013) Wolf howling
is mediated by relationship quality rather than underlying
emotional stress. Curr Biol 23:1677–1680
McConnell PB (1990) Acoustic structure and receiver response in
domestic dogs, Canis familiaris. Anim Behav 39:897–904
McConnell PB, Baylis JR (1985) Interspecific communication in
cooperative herding: acoustic and visual signals from human
shepherds and herding dogs. Z Tierpsychol 67:302–382
Mech LD (1999) Alpha status, dominance and division of labor in
wolf packs. Can J Zool 77:1196–1203
Meints K, Racca A, Hickey N (2010) Child-dog misunderstandings:
children misinterpret dogs’ facial expressions. In: Proceedings of
the 2nd Canine Science Forum, p 99
Miklósi A, Polgárdi R, Topál J, Csányi V (2000) Intentional
behaviour in dog-human communication: an experimental analysis of ‘‘showing’’ behaviour in the dog. Anim Cogn 3:159–166
Minsky M (1961) Steps toward artificial intelligence. T Ins Radio Eng
49:8–30
Molnár C, Pongrácz P, Dóka A, Miklósi A (2006) Can humans
discriminate between dogs on the base of the acoustic parameters
of barks? Behav Process 73:76–83
Molnár C, Kaplan F, Roy P, Pachet F, Pongrácz P, Dóka A, Moklósi
A (2008) Classification of dog barks: a machine learning
approach. Anim Cogn 11:389–400
Molnár C, Pongrácz P, Faragó T, Dóka A, Miklósi A (2009) Dogs
discriminate between barks: the effect of context and identity of
the caller. Behav Process 82(2):198–201
Morton E (1977) On the occurrence and significance of motivation—
structural rules in some bird and mammal sounds. Am Nat
111:855–869
Netto W, Planta D (1997) Behavioural testing for aggression in the
domestic dog. Appl Anim Behav Sci 52:243–263
Overall K, Dunham A, Frank D (2001) Frequency of nonspecific
clinical signs in dogs with separation anxiety, thunderstorm
phobia, and noise phobia, alone or in combination. J Am Vet
Med Assoc 219:467–473
Parsons S (2001) Identification of New Zeeland bats (Chalinobus
tuberculatus and Mystacina tuberculata) in flight from analysis
of echolocation calls by artificial neural networks. J Zool
253(4):447–456
Parsons S, Jones G (2000) Acoustic identification of twelve species of
echolocating bat by discriminant function analysis and artificial
neural networks. J Exp Biol 203(17):2641–2656
Pongrácz P, Molnár C, Miklósi A, Csányi V (2005) Human listeners
are able to classify dog (canis familiaris) barks recorded in
different situations. J Comp Psychol 119:136–144
Pongrácz P, Molnár C, Miklósi A (2006) Acoustic parameters of dog
barks carry emotional information for humans. Appl Anim
Behav Sci 100:228–240
Pongrácz P, Molnár C, Miklósi A (2010) Barking in family dogs: an
ethological approach. Vet J 183:141–147
Pongrácz P, Szabó E, Kis A, Péter A, Miklósi A (2014) More than
noise? Field investigations of intraspecific acoustic communication in dogs (Canis familiaris). Appl Anim Behav Sci (in press)
Quinlan R (1993) C4.5: programs for machine learning. Morgan
Kaufmann
Reid P (2009) Adapting to the human world: dogs’ responsiveness to
our social cues. Behav Process 80:325–333
Roch M, Soldevilla M, Hoenigman R, Wiggins S, Hidebrand J (2008)
Comparison of machine learning techniques for the classification
of echolocation clicks from three species of odontocetes. Can
Acoust 36(1):41–47
Root-Gutteridge H, Bencsik M, Chebli M, Gentle L, Terrell-Nield C,
Bourit A, Yarnell RW (2013) Improving individual identification
in captive Eastern grey wolves (Canis lupus lycaon) using the
time course of howl amplitudes. Bioacoustics 23(1):39–53
Saeys Y, Inza I, Larrañaga P (2007) A review of feature selection
techniques in bioinformatics. Bioinformatics 23(19):2507–2517
Serpell J, Hsu Y (2001) Development and validation of a novel
method for evaluating behavior and temperament in guide dogs.
Appl Anim Behav Sci 72:347–364
Smith A, Birnie A, Lane K, French J (2009) Production and perception of
sex differences in vocalizations of wied’s black-tufted-ear marmosets (callithrix kuhlii). Am J Primatol 71:324–332
Stone M (1974) Cross-validatory choice and assessment of statistical
predictions. J R Stat Soc B 36(2):111–147
Sucar E, Bielza C, Morales E, Hernandez-Leal P, Zaragoza J,
Larrañaga P (2014) Multi-label classification with Bayesian
network-based chain classifiers. Pattern Recogn Lett 41:14–22
Taylor A, Reby D, McComb K (2008) Human listeners attend to size
information in domestic dog growls. J Acoust Soc Am
123(5):2903–2909
Taylor A, Reby D, McComb K (2009) Context-related variation in the
vocal growling behaviour of the domestic dog (Canis familiaris).
Ethology 115(10):905–915
Taylor A, Reby D, McComb K (2010) Size communication in
domestic dog, Canis familiaris, growls. Anim Behav
79(1):205–210
Téglás E, Gergely A, Kupán K, Miklósi A, Topál J (2012) Dogs’ gaze
following is tuned to human communicative signals. Curr Biol
22:1–4
Tembrock G (1976) Canid vocalizations. Behav Process 1:57–75
Volodin I, Volodina E, Klenova A, Filatova O (2005) Individual and
sexual differences in the calls of the monomorphic white-faced
whistling duck dendrocygna viduata. Acta Ornithol 40:43–52
Wan M, Bolger N, Champagne F (2012) Human perception of fear in
dogs varies according to experience with dogs. PLoS ONE
7(e51):775
Yeon SC (2007) The vocal communication of canines. J Vet Behav
2:141–144
Yin S (2002) A new perspective on barking in dogs (Canis familiaris).
J Comp Psychol 116:189–193
Yin S, McCowan B (2004) Barking in domestic dogs: context specificity
and individual identification. Anim Behav 68:343–355
Yovel Y, Au WWL (2010) How can dolphins recognize fish
according to their echoes? A statistical analysis of fish echoes.
PLoS ONE 5(11):e14,054
123